sx1276 FSK reception non-sensitive

[moshe-braner]

Using the same hardware (T-Beam) with an sx1276, and 2 versions of the software (SoftRF), one using RadioLib and one using an older "basicmac" library, experiments show that the RadioLib version is 30dB (!) less sensitive in reception. This is in the 915 MHz band. I cannot find the reason. The sx1276 is set up with the normal beginFSK() call etc. I did notice that the LORA begin() takes a "gain" parameter, but beginFSK() does not - even though the setGain() function has a section for FSK. So I added setGain(0) to the code in SX1276::beginFSK(). Also added this line into SX1278::setGain() after the line turning on AGC when gain=0:

state |= mod->SPIsetRegValue(RADIOLIB_SX127X_REG_LNA, (1 << 5) | RADIOLIB_SX127X_LNA_BOOST_ON);

(Note: the sx1276 class is derived from the sx1278 class.) Still, no change in the rx sensitivity. What am I missing? 30dB is a lot. Maybe there is an issue with the rx/tx switching, not happening for FSK?

[moshe-braner]

Hmmm, seems like there is a typo here, although it does not explain the poor sensitivity:

SX1278::setGain(uint8_t gain) {
...
} else if(modem == RADIOLIB_SX127X_FSK_OOK) {
if(gain == 0) {
...
// gain set to 0, enable AGC loop
state |= mod->SPIsetRegValue(RADIOLIB_SX127X_REG_RX_CONFIG, RADIOLIB_SX127X_AGC_AUTO_ON, 3, 3);
} else {
state |= mod->SPIsetRegValue(RADIOLIB_SX127X_REG_RX_CONFIG, RADIOLIB_SX1278_AGC_AUTO_OFF, 3, 3);
...
}
}

• that should have been RADIOLIB_SX127X_AGC_AUTO_OFF, but both it and RADIOLIB_SX1278_AGC_AUTO_OFF are zero so this typo has no effect.

What's missing in the gain=0 case is:
mod->SPIsetRegValue(RADIOLIB_SX127X_REG_LNA, RADIOLIB_SX127X_LNA_BOOST_ON, 1, 0);
or maybe (does the nominal gain matter when AGC is on?):
mod->SPIsetRegValue(RADIOLIB_SX127X_REG_LNA, (1 << 5) | RADIOLIB_SX127X_LNA_BOOST_ON);
but I added the latter and the sensitivity improved by about 3dB (as expected from the LNA boost) but is still poor, compared to using the old library. Which did this:
writeReg(RegLna, 0x20 | 0x03); // max gain, boost enable
writeReg(FSKRegRxConfig, 0x0E); // AFC off, AGC on
Does the order of those two lines matter?

[jgromes]

This is interesting, the first question I have is how are you measuring the sensitivity?

Regarding the gain parameter, starting from version 7.7.1, it is available for all modems supported by the SX127x as a public member variable of the SX127x class. So you can do radio.gain = 4; for example.

One thing to check would be whether I have interpreted the logic of the gain parameter correctly; that is, whether gain = 1 really is the highest gain.

[moshe-braner]

The person who did the clearest test wrote this (lightly edited): " I placed my [transmitter] about 50m away with a 10dB inline attenuator and a 1/4 wave copper wire antenna directly connected. I have a short mast on my workshop with a simple 1/4 wave and groundplane antenna cabled indoors. I have various fixed and variable inline attenuators that I apply until I reach failure point (roughly where by eye 50% packets are received, the cutoff is quite sharp). Using [FSK] I consistently found: [Radiolib] failed with 12dB inline attenuation (-75 RSSI reported) [old library] failed with 41dB inline attenuation (-99 RSSI reported) ... I then swapped to [LORA], ... both [library] versions seem to behave about the same. I say about because I could not make it fail even by disconnecting the antenna cable from the input to my attenuators! Only when the pigtail was unplugged from the attenuator did it finally give up the ghost. Remarkable." Message ID: ***@***.***

…

.com>

[moshe-braner]

Maybe there is a bug in the function SX127x::calculateBWManExp(float bandwidth)? My code passes to beginFSK() the bandwidth value 234.3f (intended for a single-sided bandwidth of 125 KHz, with the modulation being +-50KHz FSK). The register RADIOLIB_SX127X_REG_RX_BW seems to be set to 0x00. (Seen via a readReg() after the setup.) The old library sets it to 0x02. 0 is not a valid value. That function should not return 0 unless the loop runs out without hitting the return((m << 3) | e);

Note that I changed this line in that function:
if(fabsf(bandwidth - ((point / 1000.0f) + 0.05f)) <= 0.5f)
to this:
if(fabsf(bandwidth - ((point * 0.001f) + 0.05f)) <= 0.5f)
because the ESP32 is very slow in FP division, but those look equivalent to me?

Or maybe I should not give it the bandwidth value 234.3? I got it from example code from somebody else. (The value 234.3 is one of the valid values for the sx1262.) The comment in SX1276.h says:
"param rxBw Receiver bandwidth in kHz. Allowed values are 2.6, 3.1, 3.9, 5.2, 6.3, 7.8, 10.4, 12.5, 15.6, 20.8, 25, 31.3, 41.7, 50, 62.5, 83.3, 100, 125, 166.7, 200 and 250 kHz."
Does that if(fabsf()) check (in an obscure way) for those values, and therefore 234.3 is not accepted and the loop runs out and the "return 0" is reached?

[moshe-braner]

BTW, the sx1262 code in RadioLib checks for the exact allowed bandwidth values, and returns an error code if invalid. But the sx1276 code only checks the range (2.6 to 250.0). Maybe should add this to SX127x::setRxBandwidth():
uint8_t regval = calculateBWManExp(rxBw);
if (regval == 0) return RADIOLIB_ERR_INVALID_RX_BANDWIDTH;
return(this->mod->SPIsetRegValue(RADIOLIB_SX127X_REG_RX_BW, regval, 4, 0));

[moshe-braner]

Experimental results: I already had the LNA boost turned on, and AGC enabled. Now just changed the rx bandwidth (passed from my app to Radiolib) from 234.3 to 250.0 KHz. So that the rx bw setting function in Radiolib can actually work, rather than setting that register to 0 which is invalid. Now I get reception down to a reported RSSI of about -100, while earlier it was failing at about -76. So that was the main issue.

I am thinking of also changing some other registers too, to make them more like the old library sets them. Those would be:

• AGC trigger on preamble detect instead of RSSI interrupt
• 1 byte preamble detect instead of 2 bytes
• disable rx auto restart
  But I am not sure what the overall effects of those changes are.

[moshe-braner]

Also noted some inconsistencies:

For the sx1276 the bandwidth passed to RadioLib is treated as single-sided. E.g., I passed 250.0 and it set the register to 1, which is 250 single-sided (table 40). Compare that to the sx1262 code, which translates 234.3 to the register value 0x0A, which AFAIK is 234.3 KHz double-sided (table 6-4)?

The sx1276 code only checked the passed bandwidth against a range, which is why I ended up with 0 in the register and poor reception. The sx1262 code checks the passed bandwidth against the exact valid values (+- 0.001), otherwise it returns an error code. The LR1110 code finds the "best fit" valid value, thus if you pass it 200.0 it will give you 234.3.

[moshe-braner]

Yeah there is a lot of FP use in RadioLib. In my modified copy of RadioLib I have converted a few to integer ops, and made other changes such as replacing FP division with multiplication, since my target processor (ESP32) is very slow doing FP division. (And thanks for not using double precision FP, which has no hardware support on the ESP32.) I realize the library keeps evolving, but I am likely to keep using the same old version for a while, to avoid re-doing my mods and to avoid breaking something. The most important modification I did was to pass a "fast" argument to begin() and beginFSK(), which causes some things in the setup to be skipped, to save time. That is because my app, in some modes, switches between RF settings (even between LORA and FSK) several times every second, so I can't have that take dozens of ms.

[moshe-braner]

Meanwhile I've made the following additional changes in the sx1276 setup, to mimic what the old library did, although these may not be necessary. They did not seem to further improve the rx sensitivity:

• turn on AFC, and trigger on preamble detect instead of RSSI interrupt - register 0D set to 1E instead of 09

• 1 byte preamble detect instead of 2 bytes - register 1F set to 8A instead of AA

• disable rx auto restart - register 27 set to 35 instead of B5 (that's for a 6-byte sync word)

Furthermore, I arranged for SX1276::beginFSK() to call setGain(0), and modified SX1278::setGain(), in FSK mode, if passed a "0", to turn on the LNA boost, as mentioned above.

In the general library, as opposed to my application, some of these things perhaps should be called separately from the user code rather than embedded within the library begin(). Although something has to be done by begin(), and may need to be redone later.

[jgromes]

FYI I pushed an update fixing this issue - thanks for finding and reporting it! In the future if there's something in your modifications you feel like would be useful in the upstream, feel free to open an issue or a new discussion.

[moshe-braner]

I see: more like the LR1110 code for setting the bandwidth, and perhaps now also the sx1262? In the sx127x code the LUT uses symbols such as "RADIOLIB_SX126X_RX_BW_2_6", I think you meant that to be "127X"?

[jgromes]

Indeed I did, and it is already fixed in e0a1e44 ;)